GFlowNet (Generative Flow Network)
A generative framework that learns to sample compositional objects proportional to a reward.
A Generative Flow Network (GFlowNet) is a probabilistic generative framework that learns policies sampling compositional objects (molecular graphs, sequences, or causal structures) with probability proportional to a given reward function. Standard generative models maximize likelihood or variational objectives. GFlowNets instead treat generation as a sequential decision-making process: an agent constructs an object step by step through actions in a directed acyclic graph of states, and training encourages the resulting flow of probability mass to satisfy a consistency condition called the flow-matching or detailed balance constraint. GFlowNets are related to reinforcement learning, but the goal differs: GFlowNets target diversity of high-reward samples rather than maximization of a single reward.
The core training objective ensures that the total flow into any intermediate state equals the total flow out, analogous to conservation laws in physical flow networks. By satisfying these constraints across all states, the learned policy generates terminal objects with frequencies proportional to their rewards. This property is useful when the reward landscape is multimodal. A greedy or maximum-likelihood approach would collapse onto a single high-reward mode; a GFlowNet explores and represents the full distribution of good solutions. Training can use variants such as trajectory balance, which give more stable and efficient gradient estimates than earlier flow-matching formulations.
GFlowNets fit scientific discovery tasks where diversity matters as much as quality. In drug discovery, a model proposing many structurally distinct high-affinity molecules is more useful than one repeatedly suggesting the same compound. They have also been applied to Bayesian structure learning, combinatorial optimization, and active learning, where maintaining uncertainty and exploring broadly is critical. Their connection to amortized variational inference and energy-based models gives them a probabilistic interpretation, allowing them to act as flexible approximate samplers for intractable posteriors.
Introduced by Yoshua Bengio and collaborators in 2021, GFlowNets have attracted research interest as a unifying framework bridging reinforcement learning, probabilistic inference, and deep generative modeling. Their ability to turn reward signals into calibrated generative distributions makes them useful wherever exploration and diversity are essential.
Research this in Signals
Signals turns a topic into a sourced research record you can inspect and rerun. Your first scan is free, and this one starts with GFlowNet (Generative Flow Network) already loaded, so edit it or scan as is.
